Automated-Portfolio, a systematic research pipeline for constructing a long-only portfolio within the S&P 500 universe, rebalanced weekly.

Asset selection starts with a factor-style scoring layer (momentum/quality) to produce a ranked candidate set. A second layer performs learning-based feature selection across accounting variables and news/sentiment features to refine the signal set and reduce redundancy, with the objective of keeping the allocation driven by a compact, stable information set rather than an over-parameterized model.

NLP is intentionally used as a conditional control mechanism instead of a direct return forecaster: sentiment signals and event intensity act as "gates" that can delay, bypass, or adjust the rebalance when the information set changes abruptly (earnings shocks, macro surprises, or narrative breaks). The end result is a Markowitz-style allocation that remains risk-aware while being less vulnerable to regime discontinuities and headline-driven distribution shifts.